Fix sloppiness in counting.

[linas]

There is some kind of thread-race condition described at lines 550-575 of link-pipeline.scm -- its talking about problems that arise when the same sentence is fed through the system in rapid succession. Apparently, the code is trying to remove Atoms to reduce clutter, but is sloppy in doing so. There is no reason to be sloppy; this should be fixed. The best way for doing this might be to use (cog-push-atomspace) and (cog-pop-atomspace)...